The ultrasensitive detection of clinically relevant somatic alterations in cancer genomes has great potential for impacting patient care, e.g. for early detection, establishing diagnoses, refining prognoses, guiding treatment, and monitoring recurrence. However, current technologies are poorly suited to the robust detection of somatic mutations present at very low frequencies (<1%). Massively parallel sequencing represents an advantageous path forward, but its sensitivity to detect very rare events is fundamentally constrained by the sequencing error rate. We have recently developed a new experimental paradigm that overcomes this limitation. In our approach, each copy of a target sequence that is present in a sample is molecularly tagged during the first cycle of a multiplex capture reaction with a unique random sequence. After amplification, target amplicons and their corresponding molecular tags are subjected to massively parallel sequencing. During analysis, the molecular tags are used to associate sequence reads sharing a common origin. Through oversampling, reads bearing the same molecular tag error-correct one another to yield an independent haploid consensus for each progenitor molecule. Furthermore, the collapsing of commonly derived reads inherently corrects for any allele-specific bias during amplification, such that estimates of mutant allele frequency can be accompanied by precise confidence bounds ( molecular counting ). Among other benefits, the approach is sensitive to at least 1 mutated sequence in a background of 10,000 unmutated copies. Here we propose the advanced development and validation of this approach for use as a clinical diagnostic. In our first aim, we will develop a multiplexed panel to broadly target common cancer associated mutations using this technology. In our second aim, we will apply the panel to the detection of minimal residual disease in acute myeloid leukemia as a prognostic marker of disease relapse. In our third aim, we will apply the panel to detection of ultra-rare mutations in circulating cell-free DNA, which is released into circulation from dying tumor cells, as a robust and non-invasive cancer diagnostic. The panel will be rigorously validated for clinical use in both aims, with performance metrics appropriately designed for the two separate analytes. The availability of robust, cost-effective, quantitative, and generically applicable tools for the ultrasensitive, multiplex detection of rare somatic events in the clinical setting will provide enhanced, transformative capabilities in the diagnosis and monitoring of cancers. The methodology will also have application to basic science cancer research.